maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/Transforms/InstSimplify/compare.ll

601 lines
12 KiB
LLVM
Raw Normal View History

At -O123 the early-cse pass is run before instcombine has run. According to my auto-simplier the transform most missed by early-cse is (zext X) != 0 -> X != 0. This patch adds this transform and some related logic to InstructionSimplify and removes some of the logic from instcombine (unfortunately not all because there are several situations in which instcombine can improve things by making new instructions, whereas instsimplify is not allowed to do this). At -O2 this often results in more than 15% more simplifications by early-cse, and results in hundreds of lines of bitcode being eliminated from the testsuite. I did see some small negative effects in the testsuite, for example a few additional instructions in three programs. One program, 483.xalancbmk, got an additional 35 instructions, which seems to be due to a function getting an additional instruction and then being inlined all over the place. llvm-svn: 123911
2011-01-20 21:21:55 +08:00
; RUN: opt < %s -instsimplify -S | FileCheck %s
target datalayout = "p:32:32"
define i1 @ptrtoint() {
; CHECK: @ptrtoint
%a = alloca i8
%tmp = ptrtoint i8* %a to i32
%r = icmp eq i32 %tmp, 0
ret i1 %r
; CHECK: ret i1 false
}
InstSimplify: Ignore pointer casts when constant folding compares between pointers. llvm-svn: 150690
2012-02-16 21:49:39 +08:00
define i1 @bitcast() {
; CHECK: @bitcast
%a = alloca i32
%b = alloca i64
%x = bitcast i32* %a to i8*
%y = bitcast i64* %b to i8*
%cmp = icmp eq i8* %x, %y
ret i1 %cmp
; CHECK-NEXT: ret i1 false
}
define i1 @gep() {
; CHECK: @gep
%a = alloca [3 x i8], align 8
%x = getelementptr inbounds [3 x i8]* %a, i32 0, i32 0
%cmp = icmp eq i8* %x, null
ret i1 %cmp
; CHECK-NEXT: ret i1 false
}
Fix a rather nasty regression from r150690: LHS != RHS does not imply LHS->stripPointerCasts() != RHS->stripPointerCasts(). llvm-svn: 150863
2012-02-18 11:29:25 +08:00
define i1 @gep2() {
; CHECK: @gep2
%a = alloca [3 x i8], align 8
%x = getelementptr inbounds [3 x i8]* %a, i32 0, i32 0
%y = getelementptr inbounds [3 x i8]* %a, i32 0, i32 0
%cmp = icmp eq i8* %x, %y
ret i1 %cmp
; CHECK-NEXT: ret i1 true
}
Reinstate the optimization from r151449 with a fix to not turn 'gep %x' into 'gep null' when the icmp predicate is unsigned (or is signed without inbounds). llvm-svn: 151467
2012-02-26 10:09:49 +08:00
; PR11238
%gept = type { i32, i32 }
@gepy = global %gept zeroinitializer, align 8
@gepz = extern_weak global %gept
define i1 @gep3() {
; CHECK: @gep3
%x = alloca %gept, align 8
%a = getelementptr %gept* %x, i64 0, i32 0
%b = getelementptr %gept* %x, i64 0, i32 1
%equal = icmp eq i32* %a, %b
ret i1 %equal
; CHECK-NEXT: ret i1 false
}
define i1 @gep4() {
; CHECK: @gep4
%x = alloca %gept, align 8
%a = getelementptr %gept* @gepy, i64 0, i32 0
%b = getelementptr %gept* @gepy, i64 0, i32 1
%equal = icmp eq i32* %a, %b
ret i1 %equal
; CHECK-NEXT: ret i1 false
}
define i1 @gep5() {
; CHECK: @gep5
%x = alloca %gept, align 8
%a = getelementptr inbounds %gept* %x, i64 0, i32 1
%b = getelementptr %gept* @gepy, i64 0, i32 0
%equal = icmp eq i32* %a, %b
ret i1 %equal
; CHECK-NEXT: ret i1 false
}
define i1 @gep6(%gept* %x) {
; Same as @gep3 but potentially null.
; CHECK: @gep6
%a = getelementptr %gept* %x, i64 0, i32 0
%b = getelementptr %gept* %x, i64 0, i32 1
%equal = icmp eq i32* %a, %b
ret i1 %equal
; CHECK-NEXT: ret i1 false
}
define i1 @gep7(%gept* %x) {
; CHECK: @gep7
%a = getelementptr %gept* %x, i64 0, i32 0
%b = getelementptr %gept* @gepz, i64 0, i32 0
%equal = icmp eq i32* %a, %b
ret i1 %equal
; CHECK: ret i1 %equal
}
define i1 @gep8(%gept* %x) {
; CHECK: @gep8
%a = getelementptr %gept* %x, i32 1
%b = getelementptr %gept* %x, i32 -1
%equal = icmp ugt %gept* %a, %b
ret i1 %equal
; CHECK: ret i1 %equal
}
Teach instsimplify how to simplify comparisons of pointers which are constant-offsets of a common base using the generic GEP-walking logic I added for computing pointer differences in the same situation. llvm-svn: 153419
2012-03-26 05:28:14 +08:00
define i1 @gep9(i8* %ptr) {
; CHECK: @gep9
; CHECK-NOT: ret
; CHECK: ret i1 true
entry:
%first1 = getelementptr inbounds i8* %ptr, i32 0
%first2 = getelementptr inbounds i8* %first1, i32 1
%first3 = getelementptr inbounds i8* %first2, i32 2
%first4 = getelementptr inbounds i8* %first3, i32 4
%last1 = getelementptr inbounds i8* %first2, i32 48
%last2 = getelementptr inbounds i8* %last1, i32 8
%last3 = getelementptr inbounds i8* %last2, i32 -4
%last4 = getelementptr inbounds i8* %last3, i32 -4
%first.int = ptrtoint i8* %first4 to i32
%last.int = ptrtoint i8* %last4 to i32
%cmp = icmp ne i32 %last.int, %first.int
ret i1 %cmp
}
define i1 @gep10(i8* %ptr) {
; CHECK: @gep10
; CHECK-NOT: ret
; CHECK: ret i1 true
entry:
%first1 = getelementptr inbounds i8* %ptr, i32 -2
%first2 = getelementptr inbounds i8* %first1, i32 44
%last1 = getelementptr inbounds i8* %ptr, i32 48
%last2 = getelementptr inbounds i8* %last1, i32 -6
%first.int = ptrtoint i8* %first2 to i32
%last.int = ptrtoint i8* %last2 to i32
%cmp = icmp eq i32 %last.int, %first.int
ret i1 %cmp
}
define i1 @gep11(i8* %ptr) {
; CHECK: @gep11
; CHECK-NOT: ret
; CHECK: ret i1 true
entry:
%first1 = getelementptr inbounds i8* %ptr, i32 -2
%last1 = getelementptr inbounds i8* %ptr, i32 48
%last2 = getelementptr inbounds i8* %last1, i32 -6
%cmp = icmp ult i8* %first1, %last2
ret i1 %cmp
}
define i1 @gep12(i8* %ptr) {
; CHECK: @gep12
; CHECK-NOT: ret
; CHECK: ret i1 %cmp
entry:
%first1 = getelementptr inbounds i8* %ptr, i32 -2
%last1 = getelementptr inbounds i8* %ptr, i32 48
%last2 = getelementptr inbounds i8* %last1, i32 -6
%cmp = icmp slt i8* %first1, %last2
ret i1 %cmp
}
At -O123 the early-cse pass is run before instcombine has run. According to my auto-simplier the transform most missed by early-cse is (zext X) != 0 -> X != 0. This patch adds this transform and some related logic to InstructionSimplify and removes some of the logic from instcombine (unfortunately not all because there are several situations in which instcombine can improve things by making new instructions, whereas instsimplify is not allowed to do this). At -O2 this often results in more than 15% more simplifications by early-cse, and results in hundreds of lines of bitcode being eliminated from the testsuite. I did see some small negative effects in the testsuite, for example a few additional instructions in three programs. One program, 483.xalancbmk, got an additional 35 instructions, which seems to be due to a function getting an additional instruction and then being inlined all over the place. llvm-svn: 123911
2011-01-20 21:21:55 +08:00
define i1 @zext(i32 %x) {
; CHECK: @zext
%e1 = zext i32 %x to i64
%e2 = zext i32 %x to i64
%r = icmp eq i64 %e1, %e2
ret i1 %r
; CHECK: ret i1 true
}
define i1 @zext2(i1 %x) {
; CHECK: @zext2
%e = zext i1 %x to i32
%c = icmp ne i32 %e, 0
ret i1 %c
; CHECK: ret i1 %x
}
According to my auto-simplifier the most common missed simplifications in optimized code are: (non-negative number)+(power-of-two) != 0 -> true and (x | 1) != 0 -> true Instcombine knows about the second one of course, but only does it if X|1 has only one use. These fire thousands of times in the testsuite. llvm-svn: 124183
2011-01-25 17:38:29 +08:00
define i1 @zext3() {
; CHECK: @zext3
%e = zext i1 1 to i32
%c = icmp ne i32 %e, 0
ret i1 %c
; CHECK: ret i1 true
}
At -O123 the early-cse pass is run before instcombine has run. According to my auto-simplier the transform most missed by early-cse is (zext X) != 0 -> X != 0. This patch adds this transform and some related logic to InstructionSimplify and removes some of the logic from instcombine (unfortunately not all because there are several situations in which instcombine can improve things by making new instructions, whereas instsimplify is not allowed to do this). At -O2 this often results in more than 15% more simplifications by early-cse, and results in hundreds of lines of bitcode being eliminated from the testsuite. I did see some small negative effects in the testsuite, for example a few additional instructions in three programs. One program, 483.xalancbmk, got an additional 35 instructions, which seems to be due to a function getting an additional instruction and then being inlined all over the place. llvm-svn: 123911
2011-01-20 21:21:55 +08:00
define i1 @sext(i32 %x) {
; CHECK: @sext
%e1 = sext i32 %x to i64
%e2 = sext i32 %x to i64
%r = icmp eq i64 %e1, %e2
ret i1 %r
; CHECK: ret i1 true
}
define i1 @sext2(i1 %x) {
; CHECK: @sext2
%e = sext i1 %x to i32
%c = icmp ne i32 %e, 0
ret i1 %c
; CHECK: ret i1 %x
}
According to my auto-simplifier the most common missed simplifications in optimized code are: (non-negative number)+(power-of-two) != 0 -> true and (x | 1) != 0 -> true Instcombine knows about the second one of course, but only does it if X|1 has only one use. These fire thousands of times in the testsuite. llvm-svn: 124183
2011-01-25 17:38:29 +08:00
define i1 @sext3() {
; CHECK: @sext3
%e = sext i1 1 to i32
%c = icmp ne i32 %e, 0
ret i1 %c
; CHECK: ret i1 true
}
define i1 @add(i32 %x, i32 %y) {
In which I discover that zero+zero is zero, d'oh! llvm-svn: 124188
2011-01-25 23:14:15 +08:00
; CHECK: @add
According to my auto-simplifier the most common missed simplifications in optimized code are: (non-negative number)+(power-of-two) != 0 -> true and (x | 1) != 0 -> true Instcombine knows about the second one of course, but only does it if X|1 has only one use. These fire thousands of times in the testsuite. llvm-svn: 124183
2011-01-25 17:38:29 +08:00
%l = lshr i32 %x, 1
In which I discover that zero+zero is zero, d'oh! llvm-svn: 124188
2011-01-25 23:14:15 +08:00
%q = lshr i32 %y, 1
%r = or i32 %q, 1
According to my auto-simplifier the most common missed simplifications in optimized code are: (non-negative number)+(power-of-two) != 0 -> true and (x | 1) != 0 -> true Instcombine knows about the second one of course, but only does it if X|1 has only one use. These fire thousands of times in the testsuite. llvm-svn: 124183
2011-01-25 17:38:29 +08:00
%s = add i32 %l, %r
%c = icmp eq i32 %s, 0
ret i1 %c
In which I discover that zero+zero is zero, d'oh! llvm-svn: 124188
2011-01-25 23:14:15 +08:00
; CHECK: ret i1 false
According to my auto-simplifier the most common missed simplifications in optimized code are: (non-negative number)+(power-of-two) != 0 -> true and (x | 1) != 0 -> true Instcombine knows about the second one of course, but only does it if X|1 has only one use. These fire thousands of times in the testsuite. llvm-svn: 124183
2011-01-25 17:38:29 +08:00
}
define i1 @add2(i8 %x, i8 %y) {
; CHECK: @add2
%l = or i8 %x, 128
%r = or i8 %y, 129
%s = add i8 %l, %r
%c = icmp eq i8 %s, 0
ret i1 %c
; CHECK: ret i1 false
}
In which I discover that zero+zero is zero, d'oh! llvm-svn: 124188
2011-01-25 23:14:15 +08:00
define i1 @add3(i8 %x, i8 %y) {
; CHECK: @add3
%l = zext i8 %x to i32
%r = zext i8 %y to i32
%s = add i32 %l, %r
%c = icmp eq i32 %s, 0
ret i1 %c
; CHECK: ret i1 %c
}
Teach instsimplify that X+Y>=X+Z is the same as Y>=Z if neither side overflows, plus some variations of this. According to my auto-simplifier this occurs a lot but usually in combination with max/min idioms. Because max/min aren't handled yet this unfortunately doesn't have much effect in the testsuite. llvm-svn: 125462
2011-02-14 01:15:40 +08:00
define i1 @add4(i32 %x, i32 %y) {
; CHECK: @add4
%z = add nsw i32 %y, 1
%s1 = add nsw i32 %x, %y
%s2 = add nsw i32 %x, %z
%c = icmp slt i32 %s1, %s2
ret i1 %c
; CHECK: ret i1 true
}
define i1 @add5(i32 %x, i32 %y) {
; CHECK: @add5
%z = add nuw i32 %y, 1
%s1 = add nuw i32 %x, %z
%s2 = add nuw i32 %x, %y
%c = icmp ugt i32 %s1, %s2
ret i1 %c
; CHECK: ret i1 true
}
According to my auto-simplifier the most common missed simplifications in optimized code are: (non-negative number)+(power-of-two) != 0 -> true and (x | 1) != 0 -> true Instcombine knows about the second one of course, but only does it if X|1 has only one use. These fire thousands of times in the testsuite. llvm-svn: 124183
2011-01-25 17:38:29 +08:00
define i1 @addpowtwo(i32 %x, i32 %y) {
; CHECK: @addpowtwo
%l = lshr i32 %x, 1
%r = shl i32 1, %y
%s = add i32 %l, %r
%c = icmp eq i32 %s, 0
ret i1 %c
; CHECK: ret i1 false
}
define i1 @or(i32 %x) {
; CHECK: @or
%o = or i32 %x, 1
%c = icmp eq i32 %o, 0
ret i1 %c
; CHECK: ret i1 false
}
Fix typo: should have been testing that X was odd, not V. llvm-svn: 124533
2011-01-29 21:27:00 +08:00
define i1 @shl(i32 %x) {
; CHECK: @shl
%s = shl i32 1, %x
%c = icmp eq i32 %s, 0
ret i1 %c
; CHECK: ret i1 false
}
Teach instruction simplify to use constant ranges to solve problems of the form "icmp pred %X, CI" and a number of examples where "%X = binop %Y, CI2". Some of these cases (div and rem) used to make it through opt -O2, but the others are probably now making code elsewhere redundant (probably instcombine). llvm-svn: 126988
2011-03-04 15:00:57 +08:00
define i1 @lshr1(i32 %x) {
; CHECK: @lshr1
Fix typo: should have been testing that X was odd, not V. llvm-svn: 124533
2011-01-29 21:27:00 +08:00
%s = lshr i32 -1, %x
%c = icmp eq i32 %s, 0
ret i1 %c
; CHECK: ret i1 false
}
Teach instruction simplify to use constant ranges to solve problems of the form "icmp pred %X, CI" and a number of examples where "%X = binop %Y, CI2". Some of these cases (div and rem) used to make it through opt -O2, but the others are probably now making code elsewhere redundant (probably instcombine). llvm-svn: 126988
2011-03-04 15:00:57 +08:00
define i1 @lshr2(i32 %x) {
; CHECK: @lshr2
%s = lshr i32 %x, 30
%c = icmp ugt i32 %s, 8
ret i1 %c
; CHECK: ret i1 false
}
define i1 @ashr1(i32 %x) {
; CHECK: @ashr1
Fix typo: should have been testing that X was odd, not V. llvm-svn: 124533
2011-01-29 21:27:00 +08:00
%s = ashr i32 -1, %x
%c = icmp eq i32 %s, 0
ret i1 %c
; CHECK: ret i1 false
}
Improve threading of comparisons over select instructions (spotted by my auto-simplifier). This has a big impact on Ada code, but not much else. Unfortunately the impact is mostly negative! This is due to PR9004 (aka SCCP failing to resolve conditional branch conditions in the destination blocks of the branch), in which simple correlated expressions are not resolved but complicated ones are, so simplifying has a bad effect! llvm-svn: 124788
2011-02-03 17:37:39 +08:00
Teach instruction simplify to use constant ranges to solve problems of the form "icmp pred %X, CI" and a number of examples where "%X = binop %Y, CI2". Some of these cases (div and rem) used to make it through opt -O2, but the others are probably now making code elsewhere redundant (probably instcombine). llvm-svn: 126988
2011-03-04 15:00:57 +08:00
define i1 @ashr2(i32 %x) {
; CHECK: @ashr2
%s = ashr i32 %x, 30
%c = icmp slt i32 %s, -5
ret i1 %c
; CHECK: ret i1 false
}
Improve threading of comparisons over select instructions (spotted by my auto-simplifier). This has a big impact on Ada code, but not much else. Unfortunately the impact is mostly negative! This is due to PR9004 (aka SCCP failing to resolve conditional branch conditions in the destination blocks of the branch), in which simple correlated expressions are not resolved but complicated ones are, so simplifying has a bad effect! llvm-svn: 124788
2011-02-03 17:37:39 +08:00
define i1 @select1(i1 %cond) {
; CHECK: @select1
%s = select i1 %cond, i32 1, i32 0
%c = icmp eq i32 %s, 1
ret i1 %c
; CHECK: ret i1 %cond
}
define i1 @select2(i1 %cond) {
; CHECK: @select2
%x = zext i1 %cond to i32
%s = select i1 %cond, i32 %x, i32 0
%c = icmp ne i32 %s, 0
ret i1 %c
; CHECK: ret i1 %cond
}
define i1 @select3(i1 %cond) {
; CHECK: @select3
%x = zext i1 %cond to i32
%s = select i1 %cond, i32 1, i32 %x
%c = icmp ne i32 %s, 0
ret i1 %c
; CHECK: ret i1 %cond
}
define i1 @select4(i1 %cond) {
; CHECK: @select4
%invert = xor i1 %cond, 1
%s = select i1 %invert, i32 0, i32 1
%c = icmp ne i32 %s, 0
ret i1 %c
; CHECK: ret i1 %cond
}
Optimize "icmp pred (urem X, Y), Y" --> true/false depending on pred. There's more work to do here, "icmp ult (urem X, 10), 11" doesn't optimize away yet. Fixes example 3 from PR9343! llvm-svn: 126741
2011-03-01 16:15:50 +08:00
Reapply commit 143214 with a fix: m_ICmp doesn't match conditions with the given predicate, it matches any condition and returns the predicate - d'oh! Original commit message: The expression icmp eq (select (icmp eq x, 0), 1, x), 0 folds to false. Spotted by my super-optimizer in 186.crafty and 450.soplex. We really need a proper infrastructure for handling generalizations of this kind of thing (which occur a lot), however this case is so simple that I decided to go ahead and implement it directly. llvm-svn: 143318
2011-10-31 03:56:36 +08:00
define i1 @select5(i32 %x) {
; CHECK: @select5
%c = icmp eq i32 %x, 0
%s = select i1 %c, i32 1, i32 %x
%c2 = icmp eq i32 %s, 0
ret i1 %c2
; CHECK: ret i1 false
}
define i1 @select6(i32 %x) {
; CHECK: @select6
%c = icmp sgt i32 %x, 0
%s = select i1 %c, i32 %x, i32 4
%c2 = icmp eq i32 %s, 0
ret i1 %c2
; CHECK: ret i1 %c2
}
Optimize "icmp pred (urem X, Y), Y" --> true/false depending on pred. There's more work to do here, "icmp ult (urem X, 10), 11" doesn't optimize away yet. Fixes example 3 from PR9343! llvm-svn: 126741
2011-03-01 16:15:50 +08:00
define i1 @urem1(i32 %X, i32 %Y) {
; CHECK: @urem1
%A = urem i32 %X, %Y
%B = icmp ult i32 %A, %Y
ret i1 %B
; CHECK: ret i1 true
}
define i1 @urem2(i32 %X, i32 %Y) {
; CHECK: @urem2
%A = urem i32 %X, %Y
%B = icmp eq i32 %A, %Y
ret i1 %B
Fix mistyped CHECK lines. llvm-svn: 127366
2011-03-10 06:07:31 +08:00
; CHECK: ret i1 false
Optimize "icmp pred (urem X, Y), Y" --> true/false depending on pred. There's more work to do here, "icmp ult (urem X, 10), 11" doesn't optimize away yet. Fixes example 3 from PR9343! llvm-svn: 126741
2011-03-01 16:15:50 +08:00
}
Teach instruction simplify to use constant ranges to solve problems of the form "icmp pred %X, CI" and a number of examples where "%X = binop %Y, CI2". Some of these cases (div and rem) used to make it through opt -O2, but the others are probably now making code elsewhere redundant (probably instcombine). llvm-svn: 126988
2011-03-04 15:00:57 +08:00
define i1 @urem3(i32 %X) {
; CHECK: @urem3
%A = urem i32 %X, 10
%B = icmp ult i32 %A, 15
ret i1 %B
; CHECK: ret i1 true
}
define i1 @urem4(i32 %X) {
; CHECK: @urem4
%A = urem i32 %X, 15
%B = icmp ult i32 %A, 10
ret i1 %B
; CHECK: ret i1 %B
}
Fold "icmp pred (srem X, Y), Y" like we do for urem. Handle signed comparisons in the urem case, though not the other way around. This is enough to get #3 from PR9343! llvm-svn: 126991
2011-03-04 18:06:52 +08:00
define i1 @urem5(i16 %X, i32 %Y) {
; CHECK: @urem5
%A = zext i16 %X to i32
%B = urem i32 %A, %Y
%C = icmp slt i32 %B, %Y
ret i1 %C
; CHECK: ret i1 true
}
Add another micro-optimization. Apologies for the lack of refactoring, but I gave up when I realized I couldn't come up with a good name for what the refactored function would be, to describe what it does. This is PR9343 test12, which is test3 with arguments reordered. Whoops! llvm-svn: 127318
2011-03-09 14:26:03 +08:00
define i1 @urem6(i32 %X, i32 %Y) {
; CHECK: @urem6
%A = urem i32 %X, %Y
%B = icmp ugt i32 %Y, %A
ret i1 %B
; CHECK: ret i1 true
}
Teach instruction simplify to use constant ranges to solve problems of the form "icmp pred %X, CI" and a number of examples where "%X = binop %Y, CI2". Some of these cases (div and rem) used to make it through opt -O2, but the others are probably now making code elsewhere redundant (probably instcombine). llvm-svn: 126988
2011-03-04 15:00:57 +08:00
define i1 @srem1(i32 %X) {
; CHECK: @srem1
%A = srem i32 %X, -5
%B = icmp sgt i32 %A, 5
ret i1 %B
; CHECK: ret i1 false
}
Teach ComputeMaskedBits about nsw on add. I don't think there's anything we can do with nuw here, but sub and mul should be given similar treatment. Fixes PR9343 #15! llvm-svn: 127463
2011-03-11 17:00:19 +08:00
; PR9343 #15
; CHECK: @srem2
; CHECK: ret i1 false
define i1 @srem2(i16 %X, i32 %Y) {
%A = zext i16 %X to i32
%B = add nsw i32 %A, 1
%C = srem i32 %B, %Y
%D = icmp slt i32 %C, 0
ret i1 %D
}
Teach ComputeMaskedBits about sub nsw. llvm-svn: 127548
2011-03-13 01:18:11 +08:00
; CHECK: @srem3
; CHECK-NEXT: ret i1 false
define i1 @srem3(i16 %X, i32 %Y) {
%A = zext i16 %X to i32
%B = or i32 2147483648, %A
%C = sub nsw i32 1, %B
%D = srem i32 %C, %Y
%E = icmp slt i32 %D, 0
ret i1 %E
}
Teach instruction simplify to use constant ranges to solve problems of the form "icmp pred %X, CI" and a number of examples where "%X = binop %Y, CI2". Some of these cases (div and rem) used to make it through opt -O2, but the others are probably now making code elsewhere redundant (probably instcombine). llvm-svn: 126988
2011-03-04 15:00:57 +08:00
define i1 @udiv1(i32 %X) {
; CHECK: @udiv1
%A = udiv i32 %X, 1000000
%B = icmp ult i32 %A, 5000
ret i1 %B
; CHECK: ret i1 true
}
Thread comparisons over udiv/sdiv/ashr/lshr exact and lshr nuw/nsw whenever possible. This goes into instcombine and instsimplify because instsimplify doesn't need to check hasOneUse since it returns (almost exclusively) constants. This fixes PR9343 #4 #5 and #8! llvm-svn: 127064
2011-03-05 13:19:11 +08:00
define i1 @udiv2(i32 %X, i32 %Y, i32 %Z) {
; CHECK: @udiv2
%A = udiv exact i32 10, %Z
%B = udiv exact i32 20, %Z
%C = icmp ult i32 %A, %B
ret i1 %C
; CHECK: ret i1 true
}
Fold icmp ugt (udiv X, Y), X to false. Spotted by my super-optimizer in 186.crafty. llvm-svn: 143209
2011-10-29 02:17:44 +08:00
define i1 @udiv3(i32 %X, i32 %Y) {
; CHECK: @udiv3
%A = udiv i32 %X, %Y
%C = icmp ugt i32 %A, %X
ret i1 %C
; CHECK: ret i1 false
}
define i1 @udiv4(i32 %X, i32 %Y) {
; CHECK: @udiv4
%A = udiv i32 %X, %Y
%C = icmp ule i32 %A, %X
ret i1 %C
; CHECK: ret i1 true
}
define i1 @udiv5(i32 %X) {
; CHECK: @udiv5
%A = udiv i32 123, %X
%C = icmp ugt i32 %A, 124
ret i1 %C
; CHECK: ret i1 false
}
Fix code to match comment. Fixes PR11340, a regression from r143209. llvm-svn: 144121
2011-11-09 05:08:02 +08:00
; PR11340
define i1 @udiv6(i32 %X) nounwind {
; CHECK: @udiv6
%A = udiv i32 1, %X
%C = icmp eq i32 %A, 0
ret i1 %C
; CHECK: ret i1 %C
}
Teach instruction simplify to use constant ranges to solve problems of the form "icmp pred %X, CI" and a number of examples where "%X = binop %Y, CI2". Some of these cases (div and rem) used to make it through opt -O2, but the others are probably now making code elsewhere redundant (probably instcombine). llvm-svn: 126988
2011-03-04 15:00:57 +08:00
define i1 @sdiv1(i32 %X) {
; CHECK: @sdiv1
%A = sdiv i32 %X, 1000000
%B = icmp slt i32 %A, 3000
ret i1 %B
; CHECK: ret i1 true
}
define i1 @or1(i32 %X) {
; CHECK: @or1
%A = or i32 %X, 62
%B = icmp ult i32 %A, 50
ret i1 %B
; CHECK: ret i1 false
}
define i1 @and1(i32 %X) {
; CHECK: @and1
%A = and i32 %X, 62
%B = icmp ugt i32 %A, 70
ret i1 %B
; CHECK: ret i1 false
}
Reapply commit 143028 with a fix: the problem was casting a ConstantExpr Mul using BinaryOperator (which only works for instructions) when it should have been a cast to OverflowingBinaryOperator (which also works for constants). While there, correct a few other dubious looking uses of BinaryOperator. Thanks to Chad Rosier for the testcase. Original commit message: My super-optimizer noticed that we weren't folding this expression to true: (x *nsw x) sgt 0, where x = (y | 1). This occurs in 464.h264ref. llvm-svn: 143125
2011-10-28 03:16:21 +08:00
define i1 @mul1(i32 %X) {
; CHECK: @mul1
; Square of a non-zero number is non-zero if there is no overflow.
%Y = or i32 %X, 1
%M = mul nuw i32 %Y, %Y
%C = icmp eq i32 %M, 0
ret i1 %C
; CHECK: ret i1 false
}
define i1 @mul2(i32 %X) {
; CHECK: @mul2
; Square of a non-zero number is positive if there is no signed overflow.
%Y = or i32 %X, 1
%M = mul nsw i32 %Y, %Y
%C = icmp sgt i32 %M, 0
ret i1 %C
; CHECK: ret i1 true
}
define i1 @mul3(i32 %X, i32 %Y) {
; CHECK: @mul3
; Product of non-negative numbers is non-negative if there is no signed overflow.
%XX = mul nsw i32 %X, %X
%YY = mul nsw i32 %Y, %Y
%M = mul nsw i32 %XX, %YY
%C = icmp sge i32 %M, 0
ret i1 %C
; CHECK: ret i1 true
}
Revert commit 149912 (lattner) and add a testcase that shows the problem (which is that patterns no longer match for vectors of booleans, because you only get ConstantDataVector when the vector element type is i8, i16, etc, not when it is i1). Original commit message: Remove some dead code and tidy things up now that vectors use ConstantDataVector instead of always using ConstantVector. llvm-svn: 150246
2012-02-10 22:26:42 +08:00
define <2 x i1> @vectorselect1(<2 x i1> %cond) {
; CHECK: @vectorselect1
%invert = xor <2 x i1> %cond, <i1 1, i1 1>
%s = select <2 x i1> %invert, <2 x i32> <i32 0, i32 0>, <2 x i32> <i32 1, i32 1>
%c = icmp ne <2 x i32> %s, <i32 0, i32 0>
ret <2 x i1> %c
; CHECK: ret <2 x i1> %cond
}
Fix PR11948: the result type of an icmp may be a vector of boolean - don't assume it is a boolean. llvm-svn: 150247
2012-02-10 22:31:24 +08:00
fold comparisons of gep'd alloca points with null to false, implementing PR12013. We now compile the testcase to: __Z4testv: ## @_Z4testv ## BB#0: ## %_ZN4llvm15SmallVectorImplIiE9push_backERKi.exit pushq %rbx subq $64, %rsp leaq 32(%rsp), %rbx movq %rbx, (%rsp) leaq 64(%rsp), %rax movq %rax, 16(%rsp) movl $1, 32(%rsp) leaq 36(%rsp), %rax movq %rax, 8(%rsp) leaq (%rsp), %rdi callq __Z1gRN4llvm11SmallVectorIiLj8EEE movq (%rsp), %rdi cmpq %rbx, %rdi je LBB0_2 ## BB#1: callq _free LBB0_2: ## %_ZN4llvm11SmallVectorIiLj8EED1Ev.exit addq $64, %rsp popq %rbx ret instead of: __Z4testv: ## @_Z4testv ## BB#0: pushq %rbx subq $64, %rsp xorl %eax, %eax leaq (%rsp), %rbx addq $32, %rbx movq %rbx, (%rsp) movq %rbx, 8(%rsp) leaq 64(%rsp), %rcx movq %rcx, 16(%rsp) je LBB0_2 ## BB#1: movl $1, 32(%rsp) movq %rbx, %rax LBB0_2: ## %_ZN4llvm15SmallVectorImplIiE9push_backERKi.exit addq $4, %rax movq %rax, 8(%rsp) leaq (%rsp), %rdi callq __Z1gRN4llvm11SmallVectorIiLj8EEE movq (%rsp), %rdi cmpq %rbx, %rdi je LBB0_4 ## BB#3: callq _free LBB0_4: ## %_ZN4llvm11SmallVectorIiLj8EED1Ev.exit addq $64, %rsp popq %rbx ret This doesn't shrink clang noticably though. llvm-svn: 150944
2012-02-20 08:42:49 +08:00
; PR11948
define <2 x i1> @vectorselectcrash(i32 %arg1) {
Fix PR11948: the result type of an icmp may be a vector of boolean - don't assume it is a boolean. llvm-svn: 150247
2012-02-10 22:31:24 +08:00
%tobool40 = icmp ne i32 %arg1, 0
%cond43 = select i1 %tobool40, <2 x i16> <i16 -5, i16 66>, <2 x i16> <i16 46, i16 1>
%cmp45 = icmp ugt <2 x i16> %cond43, <i16 73, i16 21>
ret <2 x i1> %cmp45
}
fold comparisons of gep'd alloca points with null to false, implementing PR12013. We now compile the testcase to: __Z4testv: ## @_Z4testv ## BB#0: ## %_ZN4llvm15SmallVectorImplIiE9push_backERKi.exit pushq %rbx subq $64, %rsp leaq 32(%rsp), %rbx movq %rbx, (%rsp) leaq 64(%rsp), %rax movq %rax, 16(%rsp) movl $1, 32(%rsp) leaq 36(%rsp), %rax movq %rax, 8(%rsp) leaq (%rsp), %rdi callq __Z1gRN4llvm11SmallVectorIiLj8EEE movq (%rsp), %rdi cmpq %rbx, %rdi je LBB0_2 ## BB#1: callq _free LBB0_2: ## %_ZN4llvm11SmallVectorIiLj8EED1Ev.exit addq $64, %rsp popq %rbx ret instead of: __Z4testv: ## @_Z4testv ## BB#0: pushq %rbx subq $64, %rsp xorl %eax, %eax leaq (%rsp), %rbx addq $32, %rbx movq %rbx, (%rsp) movq %rbx, 8(%rsp) leaq 64(%rsp), %rcx movq %rcx, 16(%rsp) je LBB0_2 ## BB#1: movl $1, 32(%rsp) movq %rbx, %rax LBB0_2: ## %_ZN4llvm15SmallVectorImplIiE9push_backERKi.exit addq $4, %rax movq %rax, 8(%rsp) leaq (%rsp), %rdi callq __Z1gRN4llvm11SmallVectorIiLj8EEE movq (%rsp), %rdi cmpq %rbx, %rdi je LBB0_4 ## BB#3: callq _free LBB0_4: ## %_ZN4llvm11SmallVectorIiLj8EED1Ev.exit addq $64, %rsp popq %rbx ret This doesn't shrink clang noticably though. llvm-svn: 150944
2012-02-20 08:42:49 +08:00
; PR12013
define i1 @alloca_compare(i64 %idx) {
%sv = alloca { i32, i32, [124 x i32] }
%1 = getelementptr inbounds { i32, i32, [124 x i32] }* %sv, i32 0, i32 2, i64 %idx
%2 = icmp eq i32* %1, null
ret i1 %2
; CHECK: alloca_compare
; CHECK: ret i1 false
}
fix PR12075, a regression in a recent transform I added. In unreachable code, gep chains can be infinite. Just like "stripPointerCasts", use a set to keep track of visited instructions so we don't recurse infinitely. llvm-svn: 151383
2012-02-25 03:01:58 +08:00
; PR12075
define i1 @infinite_gep() {
ret i1 1
unreachableblock:
%X = getelementptr i32 *%X, i32 1
%Y = icmp eq i32* %X, null
ret i1 %Y
}